Dark Matter
About 27% of the universe is invisible matter that shows up only through its gravity; nobody knows what it is.
Open in the interactive tree →Since Zwicky (1933) and Rubin (1970s) it is known that galaxies rotate too fast for their visible mass. The cosmic microwave background and galaxy clusters indicate about five times as much dark matter as ordinary matter. Candidates are WIMPs, axions, sterile neutrinos and primordial black holes; the alternative is modified gravity such as MOND.
As of October 2026
Direct searches have found nothing so far. LZ (South Dakota) reported 417 live days (March 2023 to April 2025, the largest dark-matter dataset so far) on 8 December 2025: no WIMP signal, world-leading limits above 5 GeV, a first search down to 3 GeV and the first detection of solar 8B neutrinos (about 4.5 sigma), so the neutrino fog is already reached at low masses. XENONnT's 3.1 tonne-year WIMP search (2025) reaches 1.7e-47 cm^2 at 30 GeV, and its 7.8 tonne-year search for light dark matter (3 to 8 GeV, 2026) found no excess either. The detectors are approaching the 'neutrino fog' where solar neutrinos mimic a signal, and no dark-matter particle has been produced at the LHC.
What is missing
- Detection of a particle, or convincing proof that none exists
- Next-generation detectors (tens of tonnes of xenon) and methods that get past the neutrino fog
- More sensitive axion searches and tests of MOND-like theories on galaxy clusters
- More data on small-scale structure (Rubin Observatory, Euclid)
Becomes possible once solved
- Knowing what 85% of all matter is made of
- New particle physics beyond the Standard Model
- More precise models of how galaxies form
Open steps
- Seeing past the neutrino fog Medium AI leverageTell dark-matter recoils from solar-neutrino recoils in tens-of-tonne xenon detectors, using statistical or directional methods.
- Faster, more sensitive axion searches Medium AI leverageScan the axion mass range with haloscopes faster and with fewer false alarms, and design new detector concepts.
- Dark-matter clumps from strong lenses High AI leverageMeasure the mass function and density slopes of small dark-matter clumps in thousands of strong lenses, to separate cold, warm, fuzzy and self-interacting models.
- Testing MOND-like gravity on clusters Medium AI leverageDecide whether modified gravity can explain galaxy-cluster, lensing and microwave-background data as well as dark matter does, with joint fits.
Where AI could help
Medium AI leverage. Lens, survey and detector data analysis speed up a lot with AI, but finding the particle needs bigger detectors and years of data.
- Infer dark-matter substructure from thousands of Euclid and Rubin lens images with neural inference
- Emulate expensive N-body and hydrodynamic simulations to test dark-matter models
- Use anomaly detection and ML background rejection in xenon and axion detectors
- Speed up fits of galaxy-cluster and rotation-curve data for MOND-like theories
Shown so far
- In August 2023 a neural likelihood-ratio method was applied to real Hubble Space Telescope strong-lens images to measure dark-matter subhalo density slopes (found steeper than cold-dark-matter predictions). source
- In March 2022 researchers showed that a simulation-based neural pipeline can infer the dark-matter subhalo mass function from populations of hundreds of strong lenses. source
Prerequisites
- Expansion & Big Bang1929
- Evidence for Dark Matter1970Galaxy rotation and cluster motions are the first evidence for dark matter
- Standard Model of Particles1973
- High-Energy Neutrino Astronomy2013IceCube searches for neutrinos from WIMP annihilation in the Sun and sets limits on them
- Gaia: 3D Map of the Milky Way2018–2022Gaia's star motions weigh the Milky Way and map its dark-matter halo